Heterogeneous Point Set Transformers for Segmentation of Multiple View Particle Detectors

The NOvA experiment, which detects neutrino particles from the NuMI beam at Fermilab, has introduced a novel point set neural network aimed at improving the segmentation of particle data. Traditionally, data processing involved a combination of clustering approaches and convolutional neural networks, but the new model operates on sparse 2D images, specifically XZ and YZ views of the detector. This innovative approach not only reduces memory usage to less than 10% of previous methods but also achieves a remarkable 96.8% AUC score, surpassing the 85.4% score obtained when processing the views independently. This development is crucial for enhancing the efficiency of data analysis in particle physics, enabling more accurate identification and matching of neutrino particles, which is essential for advancing our understanding of fundamental physics.